At the recent International Mathematical Olympiad (IMO), AI models from Google DeepMind and OpenAI achieved gold medal scores, solving five out of six problems. Google's Gemini Deep Think scored 35 out of 42, outperforming 95% of participants. OpenAI's experimental reasoning model also attained a gold-level score of 35.
Despite these achievements, five teenage participants achieved higher scores than the AI models. One participant from China scored 40 out of 42. Google DeepMind highlighted that Gemini Deep Think used 'parallel thinking' and discovered a novel proof technique. Google waited for official IMO approval before announcing the results.
These results highlight the rapid progress of AI in mathematical reasoning, with AI systems now capable of competing at a high level in challenging competitions. However, humans still maintain an edge, demonstrating superior problem-solving skills.